Generative models for local network community detection.

Area of Science:

• Network Science
• Data Mining
• Statistical Modeling

Background:

• Local community detection focuses on identifying a single community within large networks efficiently.
• Current methods often rely on ad hoc optimization, lacking a strong theoretical foundation.

Purpose of the Study:

• To develop a novel local community detection method based on generative network models.
• To leverage network uniformity assumptions for approximating unobserved network structures.

Main Methods:

• Formulated local community detection from a generative network model perspective.
• Applied local approximation techniques to variants of the stochastic block model.
• Developed probabilistic models for local community detection.

Main Results:

• The proposed method provides a principled, probabilistic approach to local community detection.
• One approximation method converges to the conductance metric in the limit.
• Experiments demonstrate competitive or superior performance against state-of-the-art algorithms on real and synthetic datasets.

Conclusions:

• The generative model approach offers a robust framework for local community detection.
• This probabilistic method provides an efficient and effective alternative to existing techniques.
• The connection to conductance highlights the theoretical grounding of the new approach.